AI 10 Pillars

The AI 10 Pillars framework maps Accenture's responsible AI principles to assessable controls, enabling organisations to evaluate AI-integrated workloads against a structured governance model.

Overview

flowchart TD
    A["AI 10 Pillars\nFramework"] --> B["Pillar 1\nAccountability"]
    A --> C["Pillar 2\nTransparency"]
    A --> D["Pillar 3\nFairness"]
    A --> E["Pillar 4\nSafety"]
    A --> F["Pillar 5\nPrivacy"]
    A --> G["Pillar 6\nHuman Oversight"]
    A --> H["Pillar 7\nSecurity"]
    A --> I["Pillar 8\nReliability"]
    A --> J["Pillar 9\nInclusion"]
    A --> K["Pillar 10\nSustainability"]

Framework ID

ai_10_pillars

Use this ID in .swao.yml or as the --framework flag:

swao assess --app <name> --framework ai_10_pillars

Pillars

ID	Pillar	Controls	Description
P1	Accountability	3	Clear ownership of AI decisions and outcomes
P2	Transparency	4	Explainable AI behaviour and decision audit trails
P3	Fairness	3	Bias detection and equitable treatment of all groups
P4	Safety	4	Prevention of harm from AI outputs and actions
P5	Privacy	3	Data minimisation and consent in AI data pipelines
P6	Human Oversight	3	Human-in-the-loop controls for high-stakes decisions
P7	Security	3	Adversarial robustness and model integrity
P8	Reliability	3	Consistent, predictable AI performance
P9	Inclusion	2	Accessible design and equitable access
P10	Sustainability	2	Environmental impact of AI infrastructure

Key Controls

P1-C1 -- AI Accountability Owner

Every AI-integrated system must designate a named accountability owner responsible for the AI system's decisions and outcomes. Evidence: organisational chart or RACI entry.

P2-C1 -- Decision Explainability

Model outputs affecting users must be accompanied by a human-readable explanation or reasoning trace. Evidence: API response schema or UI screenshot.

P3-C1 -- Bias Evaluation

Training and production data must be evaluated for demographic bias before deployment. Evidence: bias audit report or test suite output.

P4-C1 -- Harm Guardrails

AI systems must implement output filtering or refusal mechanisms for harmful content. Evidence: system prompt, filter configuration, or test results.

Running an Assessment

swao assess --app my-ai-app --framework ai_10_pillars

SWAO analyses the source path for AI integration patterns (API calls to LLM providers, model-loading code, inference endpoints) and maps findings to the 10 pillars.